The chip’s architecture enables edge devices to run sophisticated deep learning applications that could previously run only on the cloud, doing away with the key disadvantages of the current embedded processing infrastructure. Hailo addresses these issues with its holistic solution, which completely rethinks the existing pillars of computer architecture – memory, control, and compute – and incorporates a key, comprehensive Software Development Kit (SDK) co-developed with the hardware.
The Hailo-8 processor, which features up to 26 tera operations per second (TOPS), is said to outperform all other edge processors with area and power efficiency far superior to comparable solutions by a considerable order of magnitude – all at a size smaller than a penny, including the required memory. By designing an architecture that relies on the core properties of neural networks, edge devices can now run deep learning applications at full scale more efficiently, effectively, and sustainably than traditional solutions, while significantly lowering costs.
Hailo is working with leading OEMs and tier-1 automotive companies in fields such as advanced driver-assistance systems (ADAS), as well as players in industries like smart cities and smart homes, to empower smarter edge and IoT devices. These industries often require the use of high-performance cameras to perform tasks such as semantic segmentation and object detection in real time – tasks which the Hailo-8 can perform at full resolution, while consuming only a few Watts. Hailo's redesign eliminates untenable heat dissipation issues and removes the need for active cooling systems in the automotive industry.
According to preliminary results comparing Hailo-8 to Nvidia's Xavier AGX, which runs NN benchmarks such as ResNet-50, Hailo-8 consumes almost 20 times less power while performing the same tasks, the company promises.
Hailo is now sampling its chip with select partners across multiple industries, with a focus on automotive.